0

Somewhere I heard that Kolmogorov Smirnov test depends on number of parameters, is true?

If yes, so I have problem because I have used ks.test(...) in R for normal distribution with 2 parameters and for asymetric Laplac distribution with 3 parameters. And then I wolud like to compare each testing statistic $D$.

If KS test depends on number of parameters, there is some package which taken into account this fact? I have been trying found something on net but I have not been successful. Any help will be appreciated, thanks.

Waney
  • 113
  • 4
  • 2
    Numerous posts on site discuss this test and the fact that the test assumes a completely specified distribution, and address some of your other issues. For example see [here](https://stats.stackexchange.com/questions/45033/can-i-use-kolmogorov-smirnov-test-and-estimate-distribution-parameters), and [here](https://stats.stackexchange.com/questions/111693/simulation-of-ks-test-with-estimated-parameters) and [here](https://stats.stackexchange.com/questions/237779/how-can-one-compute-lilliefors-test-for-arbitrary-distributions) - and many more besides; try the search facilities. – Glen_b Mar 20 '19 at 15:20
  • 1
    After reading some of the previous discussions you may want to revise your question. – Glen_b Mar 20 '19 at 15:20
  • I have already read some of this discusion, so it means that if parameters are known the ks test works properly irrespective of number of parameters of distribution. Thanks – Waney Mar 20 '19 at 15:34
  • It would be rare for anyone to know the parameters of a Laplace distribution. More common would be that they have been *estimated* from previous data. That's not the same as knowing them! You would need a two-sample version of the K-S test in that circumstance. – whuber Mar 20 '19 at 16:20
  • @Waney yes, for example if you have hypothesized population parameters for all the parameters, it doesn't matter how many there were, the distribution is now fully specified – Glen_b Mar 20 '19 at 23:52

0 Answers0